Progress In Electromagnetics Research C, Vol. 90, 183–193, 2019

A Small Aperture Direction Finding System with Beamformingand Null Steering Capability

Sek-Meng Sow* and Tan-Huat Chio

Abstract—A Direction Finding (DF) algorithm for small aperture DF systems is proposed.Traditionally, small aperture DF systems lack beamforming capabilities and therefore require manualrotation, which may affect the Angle of Arrival (AOA) estimation accuracy. Based on CharacteristicMode (CM) Analysis, a Multi-Feed Structural Antenna (MFSA) is developed that utilizes an electricallysmall platform as a radiator. This paper chooses a Small Unmanned Aerial Vehicle (SUAV) as a designplatform. The overlap of all Individual Element Radiation Patterns (IERPs) of the proposed MFSAcovers the entire azimuth plane. In this way, beamforming and null steering of the MFSA on theazimuth plane can be achieved by linearly combining all weighted IERPs. A new method based onVector Singular Value Decomposition (SVD) is proposed to determine the weight vector of beamforming(“Sum” pattern) and null steering (“Difference” pattern) in a specific direction. Based on the “Sum-Difference” delta method, the AOA of the Radio Frequency (RF) signal source can be estimated. Asmall aperture VHF DF system with a multi-channel digital-IF receiver is developed to experimentallyverify the proposed concept. The evaluation results show that the AOA estimation RMS error is 1.55◦,and the false detection rate is significantly improved.

1. INTRODUCTION

A Radio Direction Finding (RDF) system is defined as a system that estimates the Angle of Arrival(AOA) of an incoming signal [1]. Since the First World War and World War II, it has been widely usedto provide electronic military intelligent information from the air and at sea. In addition, it was usedfor aircraft and maritime navigation before the implementation of Global Positioning System (GPS)for civilian usage in 1980’s. However, even today, there are many other modern tasks that requiredDF systems, such as search and rescue missions, wildlife tracking, location radio marker, spectrummonitoring and interference signal tracking, to name but a few.

In general, the DF system can be classified by the linear aperture size (or maximum lineardimension) of the antenna system used. Basically, there are three classes, namely small-aperture,medium-aperture, and large-aperture DF systems [1]. In [1], the linear aperture size of an antennasystem used in a small-aperture DF system should be less than 0.8λ; for medium- and large-apertureDF systems, the size of the antenna system should be “0.3λ to 2λ” and “1λ to 100λ”, respectively.For example, the Multiple Antenna System (MAS) used in small-aperture DF system can be a 2 or4 × λ/2 dipole or a full-size circular loop with diameter of λ/π (also known as Adcock Antennas [2]);a linear antenna array or a circular antenna array is used for the medium-aperture DF; and a shapedreflector or an electrically large array with directional beam pattern is used for the large-aperture DFsystem. In addition, there are several computational methods that can be used to extract the AOAinformation of the received source signal through the DF systems. These methods are Amplitude ratio,Phase Comparison (or Doppler techniques) and Time Delay [1]. However, Phase Comparison and

Received 4 December 2018, Accepted 16 January 2019, Scheduled 4 March 2019* Corresponding author: Sek-Meng Sow (tslsowsm@nus.edu.sg).The authors are with the Temasek Laboratories, National University of Singapore, Singapore.

184 Sow and Chio

Time Delay methods require the use of complex antenna systems, such as multiple-directional antennasystems or large antenna arrays. Therefore, small aperture DF systems typically use Amplitude ratioor the Watson-Watt method [1] to estimate AOA. The Watson-Watt method obtains AOA informationbased on the amplitude difference of two antennas having orthogonal radiation patterns, such as acrossed loop (or dipole) antenna. Due to the symmetrical nature in the radiation pattern, there is anambiguity of 180◦ in the estimation result. In this case, an additional omnidirectional receiving antennais required. Furthermore, the larger the amplitude ratio changes with respect to the bearing angle, thehigher the accuracy of the estimated AOA is. In order to improve the accuracy and provide a widerangular coverage, a rotation mechanism is required. It can be implemented mechanically (for example, aturntable) or electrically (for example, a phased antenna array. Again, these solutions are not feasible forsmall-aperture DF systems. Therefore, the rotation of this type of DF system is achieved by manuallyrotating the antenna system, which may introduce uncertainty in the AOA estimation. Therefore, theAOA estimation RMS error of such a DF system is usually as high as 6.1◦ [3].

Furthermore, existing MASs for small-aperture DF systems are difficult to install on portableplatforms, especially when the platform size is comparable to the operating wavelength. For example,the VHF DF system on a small Unmanned Aerial Vehicle (SUAV) has a maximum size of 2 m. In thispaper, a complete solution is proposed for such an operating environment. The proposed DF systemincludes a Multi-Feed Structural Antenna (MFSA) on SUAV as a MAS with a digital IF receiver andis described in Section 2.1 and Section 2.2. In Section 2.3, a new mathematical method is introducedto determine the required weight vectors for beamforming (“Sum”) and null steering (“Difference”) ofthe MAS radiation pattern in the specified direction. By exploiting the ability of beamforming andnull steering, the Amplitude Comparison method or the “Sum-Difference” delta method can be usedto estimate the AOA of the incoming signal. The experimental results of the proposed DF system arediscussed in Section 3. Finally, Section 4 summarizes this paper.

2. PROPOSED SMALL-APERTURE DIRECTION FINDING (DF) SYSTEM

As described in the previous section, the proposed small-aperture DF system includes a 4-feed antennasystem and a 4-channel digital-IF receiver. The hardware block diagram of the proposed DF system isshown in Fig. 1, and each subsystem is described in Section 2.1 and Section 2.2. Since the proposed DF

Figure 1. The block diagram of the proposed small-aperture DF system.

Progress In Electromagnetics Research C, Vol. 90, 2019 185

algorithm is based on the Amplitude Comparison method, it is necessary to synthesize the “Sum” and“Difference” patterns of the MAS in the specified direction. In order to synthesize the desired radiationpattern, all weighted IERPs of the MAS can be linearly combined. However, determining the weightvector required for “Sum” and “Difference” pattern based on the Least Square Error (LSE) methodrequires the use of different constraints, which may make it difficult to provide any physical justificationof the selection. To overcome this problem, a new weight vector estimation method is suggested anddescribed in Section 2.3.

2.1. Proposed Multi-Feed Antenna Design on SUAV

For the SUAV platform, in order to be able to install an antenna of comparable dimension to it, a λ/4monopole or a whip antenna is widely used, but it may protrude from the platform, as shown in Fig. 2.As such, the aerodynamic performance of the SUAV may be impaired. To overcome this problem, if thesurface of the entire platform is electrically conductive, the platform can be used as a radiator. Here,Characteristic Mode (CM) Analysis [4] is suggested as an analytical tool for designing MAS integratedin the SUAV platform.

Figure 2. An example of a SUAV with protruded antenna installed.

The basic design flow of a MAS using CM Analysis is as follows. If the appropriated CM is notidentified in step 2, the designer needs to return to step 1 and redefines the design platform, or concludesthat the selected platform is not suitable for designing the antenna based on CM Analysis.

1. Define the design platform.2. Perform CM Analysis on the defined platform.3. CM selection.4. Determine the location and type of feed based on the Selected CMs.5. Feed design.6. MAS performance verification.

The final MAS design on the selected SUAV model is a 4-feed antenna system with an Inductive CouplingElement (ICE) selected as its feeds (as shown in Fig. 3). The Individual Element Radiation Pattern(IERP) for each feed is shown in Fig. 4. For the sake of brevity, the design details of this MAS designwill not be described here and should be provided in [5].

2.2. 4-Channel Digital IF Receiver

Since the weight vector for the synthesized radiation pattern for the MAS can be any complex vector,the weighted IERPs is more easily obtained digitally than any analogue method, such as an adjustablephase shifter with a gain programmable amplifier or attenuator. The multi-channel digital IF receiver

186 Sow and Chio

Figure 3. A 4-feed antenna system on the selected SUAV platform and its Inductive Coupling Element(ICE) design.

Figure 4. Individual Element Radiation Patterns for Feed 1 ∼ 4 of the proposed MAS.

is designed to digitize the measured signal vector and then process the digitalized data through theproposed DF algorithm (see Section 3) to extract the AOA information of the measured signal. Asystem block diagram of the proposed Digital IF receiver is shown in Fig. 5. The multi-channel receiveris based on the commercially off-the-shelf (COTS) components, which consists of a 4× single channelreceiver (using the NXP chipset, consisting of SA602 [6] and SA605 [7]), a 2× local oscillator (LO) source,a 4× channel IF buffer (AD8244, [8]), a 4× channel IF amplifier (ZFL-500LN+ [9]), and multi-channelADC (AD9249-65 [10]) with data acquisition card (HSC-ADC-EVALDZ [11]).

A total of 6 single-channel receivers (labelled T1˜T3 and B1˜B3) are fabricated and evaluated basedon key parameters such as 12 dB SINAD sensitivity, total current consumption, IF amplitude imbalance,and IF phase difference. The measured results are tabulated in Table 1. Using the “T1” receiver asa reference, the IF amplitude imbalances are within ±1 dB, and the mean and standard deviation of

Progress In Electromagnetics Research C, Vol. 90, 2019 187

Figure 5. A block diagram of the 4-channel digital-IF coherent receiver.

the measured IF phase differences are from −2.3◦ to −13.1◦ and 1.0◦ to 3.7◦, respectively. The meanof the measured phase difference is known as the deterministic quantity and can be compensated bysystem calibration. However, it is impossible to eliminate random variations, which are measured asthe standard deviation, σ, of the phase difference. To understand the effect of this random variation onthe synthesized radiation pattern, a statistical study is performed by manually introducing simulatedphase noise to the weight vector. A synthesized radiation pattern obtained by linearly combining fourIERPs with a weight vector, �w, of {0◦, 180◦, 0◦, 180◦} is used for this study. By adding phase noise toeach phase setting of �w based on the Monte Carlo Simulation method [12], the observed maximum peakgain variations when σ = 3◦ and σ = 5◦ are 0.08 dB and 0.37 dB, respectively. In general, variationsin the peak gain of the measured radiation pattern of less than 0.1 dB are considered to be negligible.Therefore, the standard deviation of the phase difference < 3◦ should be acceptable. As such, all singlechannel receivers except T2 are qualified to be used for the proposed multi-channel Digital IF receiver(as shown in Fig. 6).

Table 1. Test results for the 6 single-channel receivers.

Receiver Current 12 dB SINAD IF output (w.r.t. TI)Amp Imbalance Phase Difference

mA dBm dB Mean (◦) Std Dev (◦)T1 7.4 −104.0 - - -T2 7.4 −104.0 −0.3 −13.1 3.7T3 7.4 −104.0 0.7 −4.7 2.1B1 7.5 −105.0 0.3 −10.5 2.1B2 7.4 −104.3 0.1 −2.3 1.6B3 7.3 −104.3 0.2 −3.9 1.0

188 Sow and Chio

Data Acquisition Card Top view Side view

DC Power Supply

16-channel ADC IF Amp IF Butter

4-channel FM Receiver

Figure 6. The complete 4-channel Digital-IF coherent receiver.

2.3. Proposed Weight Vector Estimation Method

Since the resulting radiation pattern of a MAS can be obtained by linearly combining all weightedIERPs, the “Sum” (beamforming) and “Difference” (null steering) pattern for any given direction canbe obtained if the required weight vectors are known. Generally, the required weight vector, �w, isdetermined by solving the vector equation stated as Eq. (1) with some constraints, where �IERP is thepattern vector, and F is the resulting radiation pattern. This is also known as the Least Square Error(LSE) method.

F (�r,θ) =−−−−→IERP (�r, θ) �w (1)

The LSE method is well suited for determining the weight vector, �wsum, to synthesize the “Sum” patternin the specified direction, θ, which can be done by simply defining a fairly large value for F . However,this is not the case of determining the weight vector, �wdiff , for null steering by simply set F = 0. Bydoing so, it can result in �wdiff to be a null vector. In order to prevent yielding of a null vector as aresult, at least one element of �wdiff needs to be specified as a constant value. Theoretically, there shouldbe an infinite way to define constraints for determining the weight vector using the LSE method. Itis also difficult to provide a physical explanation for the constraints. Furthermore, two mathematicaloperations are required to determine the weight vectors for “Sum” and “Difference” patterns in thespecified direction, θ.

Here, a vector Singular Value Decomposition (SVD) method is proposed for determining all requiredweight vectors in a single operation. Physically, the weight vector, �wsum, that maximizes the field (orthe “Sum” pattern) in the specified direction, θ, can be considered as a complex conjugate of the patternvector, �IERP (θ). On the other hand, the weight vector, �wdiff , for the null steering (or “Difference”pattern) can be viewed as a vector orthogonal to �IERP (θ). Thus, by performing the vector SVD on thepattern vector, �IERP (θ), one weight vector of the “Sum” pattern and (n−1) orthogonal weight vectorsof the “Difference” pattern in θ direction can be obtained. The formula of the vector SVD method isstated in Eq. (2), where �wsum = �v1 and �wdiff = �vi for i = 2 to n.

−−−−→IERP (θ)1×n = u�S1×n [V]Tn×n

= [ λ1 0 . . . 0 ]1×n [ �v1 �v2 . . . �vn ]Tn×n (2)

In Eq. (2), it can be seen that (n−1) orthogonal �wdiff vectors are obtained by using the vector SVDmethod. By using all of these �wdiff to synthesize multiple “Difference” patterns of the MAS in a specified

Progress In Electromagnetics Research C, Vol. 90, 2019 189

direction based on the measured signal vector, it should be less susceptible to noise. In Section 3, thisis verified by comparing the AOA estimation measurements obtained by the two methods, the LSEmethod and the vector SVD method.

3. EXPERIMENTAL RESULTS DISCUSSION

Based on the hardware description provided in Section 2, a VHF DF system for the SUAV prototype isconstructed, as shown in Fig. 7. For measurement purpose, it can be seen in Fig. 7 that the entire DFsystem is placed on a wooden rotatable platform. The DF test setup is depicted in Fig. 8. The signalsource is located 6 m away from the Unit Under Test (UUT), which is equivalent to 2.3λ@115 MHz(where the far field radiation region of 115 MHz should be > 1.2λ). This setup can be used to emulatesignal sources from different directions by rotating the UUT. Thus, the IERP of the MAS can bemeasured. Due to the limitations of the test facility, these IERPs are only measured on the azimuthplane. Therefore, the following discussion is limited to 1D AOA estimation on the azimuth plane.However, this should not limit the proposed DF algorithm. In general, if 3D IERPs are available, itcan also be used for 2D AOA estimation applications. A MATLAB code for this algorithm has beendeveloped to estimate the AOA and control the turntable.

Front View Back View

AUT

LO2

DBF Receiver

LO1

DC PowerSource

Laptop

Figure 7. A VHF DF system for a SUAV prototype.

The proposed DF algorithm consists of two phases, a Calibration Phase and an AOA EstimationPhase. A flow chart of the proposed algorithm is shown in Fig. 9. During the Calibration Phase, theIERP of all feeds in the MAS is measured, as shown in Fig. 10. Based on these IERPs, �wsum and�wdiff are computed based on the selected method (LSE method or vector SVD method). Once all therequired weight vectors have been computed for the entire azimuth plane, the DF system is consideredto be calibrated and should be ready for AOA estimation. For verification purposes, the UUT is rotatedin the azimuth plane from −180◦ to 180◦ in 1◦ step (where the nose of the SUAV is referred as 0◦ inthe azimuth plane), and the DF system measures the signal received by the MAS, then based on the“Sum-Difference” delta method extracts the embedded AOA information. An example screen shot ofan AOA estimation with a signal source at 0◦ is shown in Fig. 11.

Basically, the performance of the proposed VHF digital DF system is evaluated based on thefollowing criteria,1) the overall root mean square (RMS) error of the estimated AOAs,2) the mean error value of the estimated AOAs,3) the standard deviation or the spread of the distribution of estimated AOAs, and

190 Sow and Chio

Figure 8. The Direction Finding test setup.

Figure 9. The process flow chart of the proposed DF algorithm.

4) the false-detection rate.Here, the overall RMS error is defined as the RMS average of all estimated AOA values over the

selected azimuth range [13], and “false-detection” is defined as the occurrence of an estimation error> 10◦. To compare the performance of two different weight vector determination methods, the LSEmethod and vector SVD method, each method collects a total of 35 sets of AOA estimation data.Statistical processing is performed on these data. For LSE and vector SVD, statistical results (suchas rms, mean, and standard deviation) of AOA estimation error and false detection rate are shown inFig. 12 and Fig. 13, respectively.

In Fig. 12 and Fig. 13, it can be seen that in the boresight, i.e., azimuth angle = 0◦, the meanand standard deviation of the AOA estimation errors of the two methods are less than 1◦. However,

Progress In Electromagnetics Research C, Vol. 90, 2019 191

Figure 10. Measured IERPs for all feed elements of the proposed MAS.

Figure 11. A computer screenshot displays the result of an estimated AOA with reference to the UUT.

Figure 12. The rms, mean and standard deviation of AOA estimation error and the false-detection rateover the entire azimuth plane based on 35 sets of estimated AOA data based on the weight determinedby LSE method.

192 Sow and Chio

Figure 13. The rms, mean and standard deviation of AOA estimation error and the false-detection rateover the entire azimuth plane based on 35 sets of estimated AOA data based on the weight determinedby Vector SVD method.

for the LSE method and SVD method, the overall RMS error values are computed to be 2.67◦ and1.55◦, respectively. This indicates that the performance of the latter over the entire azimuthal plane is42% better than the former. This also shows that the “Difference” pattern synthesized using multipleorthogonal weight vectors is less susceptible to phase noise variations in the measured signal vector.In other words, the SVD method provides a higher possibility of synthesizing a “Difference” patternwith “deeper” null even under the influence of system noise. As such, it has a lower false-detectionrate than the LSE method. In general, the maximum false-detection rate recorded is less than 12%, sothe performance of the proposed DF system should be considered acceptable. Furthermore, given theoverall RMS estimation error of 1.55◦, the proposed small-aperture DF system is qualified as a class BDF system [14], where most of the commercial small aperture DF systems (using 5-element small basedantenna system with Amplitude Comparison techniques) are classified as class D DF systems becausetheir typical overall RMS estimation error is 6.1◦ [3, 15].

4. CONCLUSIONS

The VHF DF system on the SUAV prototype is developed to experimentally verify the performanceof the proposed MAS and the proposed weight vector estimation method in Direction Finding (DF)applications. The MAS is designed based on Characteristic Mode (CM) Analysis, and the vector SVDmethod is proposed to derive all required weight vectors for the synthesis of one “Sum” and multiple“Difference” patterns. Overall, the AOA estimation performance of the proposed system indicates thatthe vector SVD method has a lower rms estimation error and a significant improvement in the falsedetection rate compared to the LSE method.

ACKNOWLEDGMENT

Here, the authors would like to thank Mr. Tan Peng Khiang and Mr. Dylan Ang, members of theantenna group technical support team of Temasek Laboratories @ National University of Singapore, fortheir innovative ideas and suggestions during the prototyping process.

REFERENCES

1. Jenkins, H. H., Small-aperture Radio Direction-finding, Artech House on Demand, 1991.2. Kratschmer, G., “Introduction into theory of direction finding,” Radiomonitoring and

Radiolocation, Vol. 2011, 2011, 2010.

Progress In Electromagnetics Research C, Vol. 90, 2019 193

3. Bullock, L. G., G. Oeh, and J. J. Sparagna, “An analysis of wide-band microwave monopulsedirection-finding techniques,” IEEE Transactions on Aerospace and Electronic Systems, Vol. 7,No. 1, 188–203, 1971.

4. Chen, Y. and C.-F. Wang, Characteristics Modes: Theory and Applications in AntennaEngineering, (No. Book, Whole), John Wiley and Sons, Inc, Hoboken, New Jersey, 2015.

5. Sow, S.-M., L. Guo, S.-G. Zhou, and T.-H. Chio, “Electrically small structural antenna designfor small UAV based on characteristics modes,” 2017 11th European Conference on Antennas andPropagation (EUCAP), 2134–2138, Paris, France, 2017.

6. SA602AN, 2018, Available: http://www.nxp.com/products/rf/rf-mixers/low-power-fm-if-systems/double-balanced-mixer-and-oscillator:SA602AN.

7. SA605, 2018, Available: https://www.nxp.com/products/rf/rf-mixers/low-power-fm-if-systems/high-performance-low-power-mixer-fm-if-system:SA605.

8. Analog Device, Single Supply, Low Power, Precision FET Input Quad Buffer, Available:http://www.analog.com/media/en/technical-documentation/data-sheets/ad8244.pdf.

9. Mini Circuits, Low Noise Amplifier, ZFL-500LN, Available: https://www.minicircuits.com/pdfs/ZFL-500LN.pdf.

10. Analog Device, “AD9249, 16 Channel, 14-Bit, 65 MSPS, Serial LVDS, 1.8 V ADC,” ed.11. AnalogDevice. HSC-ADC-EVALDZ, FPGA-Based Data Capture Kit, Available: http://www.

analog.com/en/design-center/evaluation-hardware-and-software/evaluation-boards-kits/eval-hsc-adc-evaldz.html#eb-overview.

12. Mooney, C. Z., “Monte Carlo Simulation,” Thousand Oaks, California, 2011, [Online], Available:http://methods.sagepub.com/book/monte-carlo-simulation, Accessed on 2018/06/15.

13. Recommendation ITU-R SM.2060-0 Test Procedure for Measuring Direction Finder Accuracy, 2014.14. Recommendation ITU-R SM.854-3 Direction finding and location determination at monitoring

stations, 2011.15. Kuiper, R., “Direction finding for unmanned aerial systems using rhombic antennas and amplitude

comparison monopulse,” Carleton University, 2018.